In real applications, the most commonly used regression and classification models are linear. Currently various nonlin- ear methods, such as generalized additive models (GAM), classification and regression tree (CART), multivariate adaptive regression splines (MARS) and neural network models have become popular. Among these popular non- linear methods, neural network models are attractive in their flexibility, and achieve comparable performance in prediction.
Analygen DSS incorporates nonlinear multiple regression, nonlinear logistic regression, and nonlinear multinomial logistic regression using neural networks to perform the required statistical analysis of complex systems Any nonlinear models in statistical methodology need some sort of assumptions about either distributions and/or prior knowledge for suitable functions to be used in modeling the nonlinearity, but in real applications (particularly in modeling the phenomena in market research and decision support) we actually never know which suitable functions to use. One major benefit of neural networks is their flexibility. As a consequence, in many applications, neural networks have shown better prediction ability compared to classical statistical methods.